Start Up No.1,007: should OpenAI release GPT2?, a new ‘Oumuamua theory, where flat earthers come from, and more

The US FDA is warning old folk not to be vampires. Yup. CC-licensed photo by Abc Abc on Flickr.

You can sign up to receive each day’s Start Up post by email. You’ll need to click a confirmation link, so no spam.

A selection of 10 links for you. Almost an armful. I’m @charlesarthur on Twitter. Observations and links welcome.

Blood of the young won’t spare rich old people from sadness and death, FDA says • Ars Technica

Beth Mole:


The US Food and Drug Administration issued an alert Tuesday, February 19, warning older consumers against seeking infusions of blood plasma harvested from younger people. Despite being peddled as anti-aging treatments and cures for a range of conditions, the transfusions are unproven and potentially harmful.

In a statement, FDA Commissioner Scott Gottlieb and the director of FDA’s Center for Biologics Evaluation and Research, Peter Marks, wrote: “Simply put, we’re concerned that some patients are being preyed upon by unscrupulous actors touting treatments of plasma from young donors as cures and remedies.”

Establishments in several states are now selling young blood plasma, which is the liquid portion of blood that contains proteins for clotting. The sellers suggest that doses of young plasma can treat conditions ranging from normal aging and memory loss to dementia, Parkinson’s disease, multiple sclerosis, Alzheimer’s disease, heart disease, or post-traumatic stress disorder, according to the FDA.

The claims are wild extrapolations from intriguing but preliminary findings in mouse studies.


Ah, mouse studies. When I wrote about science all the time, the phrase “shown in mice” always meant “won’t be shown in humans”. Even so, 2019 really is screwed up if one of its official warnings has to be “don’t be a vampire”.
link to this extract

Dear OpenAI: please open source your language model • The Gradient

Hugh Zhang is a research in the neuro-linguistic programming group at Stanford University:


While OpenAI is correct to be concerned about potential misuse, I disagree with their decision not to open source the GPT-2. To justify this, I first argue that only certain types of dangerous technology should be controlled by suppressing access. Then, on the basis of this analysis, I argue that withholding the full GPT-2 model is both unnecessary for safety reasons and detrimental to future progress in AI.

I broadly classify modern technology with potential for misuse as either destructive or deceptive technology. Destructive technologies operate primarily in the physical realm. Think chemical weapons, lab-engineered super viruses, lethal autonomous weapons, or the atom bomb.

On the other hand, deceptive technologies operate primarily in the realm of our minds and can potentially be misused to manipulate and control people on a broad scale. Think deepfakes, Photoshop, or looking back in history, the Internet or the printing press. With the notable exception of autonomous weapons, fears around AI misuse tend to fall into this category…

…with deceptive technologies, there is an alternative, more effective option. Instead of suppressing a technology, make knowledge of its power as public as possible. While counterintuitive, this method of control relies on realizing that deceptive technologies lose most of their powers if the public is broadly aware of the potential for manipulation. While knowledge of nuclear weapons will not save me from fallout, awareness of recent advances in speech synthesis will make me significantly more skeptical that Obama can speak Chinese. Bullets do not discriminate on one’s beliefs, but my knowledge of modern photo editing makes it difficult to convince me that Putin is capable of riding a bear.


I think OpenAI are trying to work out how to make sure it can’t be destructive.
link to this extract

No, ‘Oumuamua is not an alien spaceship. It might be even weirder • SYFY WIRE

Phil Plait:


for the math to work out with the acceleration seen, ‘Oumuamua had to be flat. Like, really flat: So thin that it looked more like a solar sail, a very thin sheet of material designed to catch sunlight and accelerate. But that, in turn, meant that ‘Oumuamua was artificial. As in, a spaceship.

Besides the obvious (it seems like a big leap!), I have my problems with this idea. Not much has changed with that hypothesis since I wrote that, and while I wouldn’t dismiss it being an alien probe out of hand, the evidence doesn’t support that conclusion, and in fact points against it.

So, I ask again: what the frak is ‘Oumuamua?

A new paper has come out that might have a solution, and it’s really clever. Maybe ‘Oumuamua’s not flat. Maybe it’s fluffy.

Fractal structures can look solid but actually be mostly empty space; this is called a Koch Curve, and snowflakes can have structures similar to this. Credit: Eric Baird / Wikimedia

When the astronomers speculated it might be a thin and flat, giving it a large area like a sail, they had to assume a density for it. That’s because the amount of pressure sunlight exerts is very small, so if an object is massive it has to be spread out very thin and big to catch enough sunlight to accelerate it enough to match the observations. So they assumed it had some normal density like 1 – 3 grams per cubic centimeter (roughly somewhere between the density of water to rock).

The new paper turns that around. Instead of assuming a density to find the area, let’s assume the size determined using normal methods is correct, use that to get an area, and from there get the density needed to match the observations.


In essence, a three-dimensional interstellar snowflake, an accretion disc from the galaxy’s early days
link to this extract

Why women are underrepresented in clinical trials • Endpoints (a science publication by Elysium Health)


Up until the late 1970s , the decades-long Baltimore Longitudinal Study of Aging, one of the world’s longest running studies of aging, followed more than 1,000 men and zero women—even though women represented the majority of the elderly population. The Physicians Health Study, which concluded in 1989 that taking low-dose aspirin might lower your risk for heart disease, included 22,000 men and zero women. And just a few years ago, researchers investigating the possible interactions between libido-boosting drug flibanserin — known as “female Viagra” — and alcohol, used a study group of 25 participants, of which twenty-three were men.

Historically, excluding women in clinical trials has been less about bias, and more so due to a lack of knowledge about the biological differences between men and women and how disease symptoms might present differently based on sex, says Dr. Natalie DiPietro Mager, an associate professor of pharmacy practice at Ohio Northern University and co-author of a recent paper documenting women’s involvement in clinical trials.

While we’ve seen a dramatic shift toward inclusion in the last 25 years, the biomedical research world still has a long way to go in terms of robust representation of women, says Mager.


(Thanks @Reynolds for the link.)
link to this extract

Interview: David Runciman • E-International Relations

Runciman is professor of politics at the University of Cambridge:


Q: You are currently engaged in two fascinating research projects, Conspiracy and Democracy and AI: Trust and Society. Are there conspiracy theories surrounding artificial intelligence in democratic societies and what impact are they having?

DR: One part of the Conspiracy and Democracy project was to look at a common view that AI (and digital technology more generally) is driving conspiracy theories because it’s allowing them to spread. There is a perspective that we live in a world where there are more and more conspiracy theories because the Internet has provided people with an opportunity to believe the craziest things. They then find other people who believe them, creating a network of people. What we’ve found is that this is probably not true. That behaviour does exist, but there are also more people who debunk conspiracy theories, so it’s not all skewed one way. Conspiracy theories are actually debunked quicker. There have always been conspiracy theories, so it’s not a phenomenon of the internet age, it goes way back.

It’s also interesting that there is a perception that this technology is responsible for creating conspiracy theories. However, it isn’t until very recently that there were conspiracy theories about the technology itself. There are lots of conspiracy theories now that say that Facebook’s Mark Zuckerberg is secretly up to something. A lot of the reporting on Cambridge Analytica is written in the style of a conspiracy theory. It involves the idea that you uncover the network that involves the Mercers, Trump, and Brexit and everything connects – they’re joining all of the dots. There are more people seeing the world and how everything connects to everything else, and this knowledge is power for a few players in the tech space. But those are still not the dominant conspiracy theories. There are many more conspiracy theories that are depressingly old-fashioned ones and they are anti-Semitic. Even some of the conspiracy theories about technology are anti-Semitic because every conspiracy theory has an anti-Semitic variant of it. It’s not particularly the dominant mode, but I suspect that it will grow.


Runciman is always worth listening to; the Talking Politics podcast, which he presents every week, is a necessary antidote to the thin gruel of analysis in most news outlets.
link to this extract

Researchers blame YouTube for the rise in Flat Earthers • Engadget

Rachel England:


Despite steps taken to counteract problematic material YouTube is still a hotbed of hoaxes and fake news — a problem that’s become so prevalent the site recently announced it is changing its AI in a bid to improve matters. But now the scope of the problem has really come to light, as new research suggests that the increasing number of Flat Earthers can be attributed to conspiracy videos hosted on the site.

According to Asheley Landrum, assistant professor of science communication at Texas Tech University, all but one of 30 Flat Earthers interviewed said they hadn’t considered the Earth to be flat until watching videos promoting the theory on YouTube. Presenting her results at the annual meeting of the American Association for the Advancement of Science in Washington DC, Landrum said that most were recommended videos after watching clips about other conspiracies, such as alternative 9/11 theories and fake moon landing theories.

Landrum’s interest in the topic was first piqued after she attended the world’s largest gatherings of Flat Earthers at the movement’s annual conference in Rayleigh, North Carolina, in 2017. She visited the conference again in 2018, when it took place in Denver, Colorado, to interview a number of the attendees. According to Landrum, the one person who didn’t point to YouTube as a catalyst for their opinion change had their mind changed by family members who themselves were convinced by YouTube videos.


link to this extract

Japan smartphone market Q42018 • Canalys Newsroom


Smartphone shipments fell 3.8% year on year in Japan to 9.9m in Q4 2018, marking a fourth consecutive quarter of shipment decline. In terms of shipment numbers, Japan came fourth worldwide, behind China, the US and India; 32.5m smartphones shipped in Japan in the whole of 2018, 1.9% fewer than in 2017.


Really wouldn’t have expected Japan to be a bigger market than, well, so many others. (Population of 126 million, 11th largest in the world; China, India and US are the three largest.)

link to this extract

America’s signature mode of transportation is high-cost rail • Hmm Daily

Jacob Bacharach:


This demonstrable ability of other nations to build public works at reasonable—or at least bearable—cost puts lie to the most common excuses for the cost and complexity of American transit projects. America is densely populated and right-of-way property is expensive here? Same goes for Western Europe. America is big? So is China. Unions? Europe, again. You can easily believe that the unique combination of these exacerbating factors of cost and distance would lead to a state where American rail would be the most costly in the world. But more costly by an order of magnitude?

American infrastructure is this costly because of immense, endemic, universal public-private corruption—systems of both direct and financialized graft at every stage of infrastructure development, from the planning to the ribbon-cutting to the use of deferred maintenance to ransack public transportation budgets for cash, year after year, after which the responsible authorities claim that fixing the century-old signals is just too damn pricey. This system of legal fraud begins with the bevies of project consultants, continues through ludicrous private contractor and labor costs, and continues when, years later, high-paid administrative fixers and new armies of consultants and contractors arrive to fix what broke because it was never maintained. It is a system of tolerated kleptocracy that may be the only thing that America still does better than anyone else in the world. It is baked into every assumption about building for the public benefit.

You will of course notice that the responsible scolds saying of a 400-mile railroad, It cannot be done, were not so much in evidence when TransCanada and ConocoPhillips, for example, ran the Keystone Pipeline across 2,000-plus unfriendly miles.


I liked his description of Megan McArdle as “a self-styled contrarian libertarian who writes well-actually columns for the Washington Post in the style of a ‘For Dummies’ book”.

Which reminds me, isn’t the US due another infrastructure week real soon now?
link to this extract

Emoji are showing up in court cases exponentially, and courts aren’t prepared • The Verge

Dami Lee:


Bay Area prosecutors were trying to prove that a man arrested during a prostitution sting was guilty of pimping charges, and among the evidence was a series of Instagram DMs he’d allegedly sent to a woman. One read: “Teamwork make the dream work” with high heels and money bag emoji placed at the end. Prosecutors said the message implied a working relationship between the two of them. The defendant said it could mean he was trying to strike up a romantic relationship. Who was right?

Emoji are showing up as evidence in court more frequently with each passing year. Between 2004 and 2019, there was an exponential rise in emoji and emoticon references in US court opinions, with over 30% of all cases appearing in 2018, according to Santa Clara University law professor Eric Goldman, who has been tracking all of the references to “emoji” and “emoticon” that show up in US court opinions. So far, the emoji and emoticons have rarely been important enough to sway the direction of a case, but as they become more common, the ambiguity in how emoji are displayed and what we interpret emoji to mean could become a larger issue for courts to contend with.

Emoticons started appearing in court in 2004, and they have since been found most commonly in sexual predation cases. But that’s just counting the cases that were able to be tracked with the words “emoji” and “emoticon.” Electronic databases of court opinions aren’t set up to handle the actual emoji, and they aren’t displayed in case database services like Westlaw or Lexis, which is where Goldman finds his references.


At which point the question is: how do you translate from emoji to legal English?
link to this extract

Morgan Stanley: Spaceflight Industries disrupting rocket launch market • CNBC

Michael Sheetz:


Spaceflight Industries has two businesses: The all-in-one launch services unit, known as Spaceflight, and a satellite imagery unit called BlackSky. The former “has launched 210 satellites” to date, Morgan Stanley said. Spaceflight isn’t slowing down, either, with contracts to launch about 100 satellites this year.

BlackSky represents the company’s reach into satellite operations. The unit successfully launched two satellites at the end of last year, Global-1 and Global-2, and expects to launch six more this year. Spaceflight Industries aims to eventually have constellation of 60 satellites to provide high-resolution photos of Earth nearly on demand.

The company announced a $150m fundraising round in March for the first 20 satellites of the BlackSky constellation.

Additionally, BlackSky is one of several companies working with Amazon Web Services for the recently-announced AWS Ground Station business. Amazon’s cloud business is building a network of satellite connection facilities, representing the e-commerce giant’s first public move into space-related hardware.

Ground stations are a vital link for transmitting data to-and-from satellites in orbit, used by companies engaged in a variety of activities like weather forecasting, communications and broadcasting. AWS Ground Station aims to remove the heavy capital costs for these companies of building their own ground station networks off of satellite operators.


The element of this I find jawdropping is AWS Ground Station, whose announcement completely passed me by. It is “a fully managed service that lets you control satellite communications, downlink and process satellite data, and scale your satellite operations quickly, easily and cost-effectively without having to worry about building or managing your own ground station infrastructure.” Wow.
link to this extract

Errata, corrigenda and ai no corrida: pushups by firefighters (in yesterday’s post) probably aren’t a good guide to how the general population is going to get on or a prediction for heart disease – especially as women show different symptoms from men, and few of the firefighters were men. Thanks @Reynolds for pointing this out.

6 thoughts on “Start Up No.1,007: should OpenAI release GPT2?, a new ‘Oumuamua theory, where flat earthers come from, and more

  1. In other news, Xiaomi launched its mainstream flagship today. Decidedly moving upmarket: there are versions with an iridescent back !
    Still full of those weird flagship limitations (no SD slot, no audio jack, no FM radio, meh battery), but at least the notch is small this time around, they’ve stopped overcompensating for having nothing worthy to put in it.
    It’s the 1st SD855 phone, the price is OK, and as usual with Xiaomi no surprise problems.

    The Pocophone version of it that’s bound to happen, with a nasty plastic casing, so-so screen and camera, but the same innards and those extra low-ranger features will probably be more interesting for the non-aesthetes.

    A Redmi Note still makes a lot more sense unless you really want wireless charging, better pics especially in low light, more performance, or AMOLED; and can do w/o the low-ranger features.

  2. I’m confused by Apple’s move to run iOS apps on Macs.

    To me it makes perfect sense since the PC and the Tablet form factor and use cases and screen format aren’t that far apart, certainly closer than Tablet and Phone which have run the same apps since day one. And Android and Windows and ChromeOS have had competent mixed-mode Touch & Keyboard+Pointer input for years.

    But if the apps can be the same, why not make touch Macs and pointer iPads ? Apple is saying the apps can support both, and the OS itself (main UI, tools/utils, 1st-party apps) is just a series of apps too. The only explanation I can come up with is that they want to sell several devices where one could suffice. The extra cost+weight of a touchscreen on Macs or the slight effort to support a mouse on iOS don’t seem enough reason. It’s not toaster-fridge. It’s oven-microwave.

  3. Not sure if this is good or bad. Faster is better, but proprietary is worse, and I think AMP is, for only fake-FOSS, like most Google stuff.

  4. Ah well, Samsung’s Flex phone is out, $2k for a 7.3″ screen. Looks about 4:3 format which is nice.

    My 7″ Mi Max cost me $250 but I must take it out of my pocket before sitting down, and it has a sucky 18:9 format. Also less powerful and worse pics, but I don’t really care.

    I’m really curious if that thing will sell. And how well it actually works and withstands time and bumps.

  5. Aaaaand the S10. Nothing much to notice. On the good side, they haven’t caught flagship-unfeaturitis, so have an SD slot and an audio jack (usually means FM radio too but that’s unconfirmed), even a large battery. On the negative side, the punch hole front cameras are very far from the edge, so there’s massive wasted space at the top of the screen, about 2x a normal notification area. The ultra-bright screen is nice, though I never understood why reflectance is never mentioned, I’m kind of tired of seeing my face when I’m watching a movie.
    I’m not sure anyone cares much about HDR10+ (many people use a screen protector which throws those nice screens completely out of whack anyway) or where the touchID sensor is. Oh well, bragging rights.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.